-
Notifications
You must be signed in to change notification settings - Fork 0
/
colorizer_nongan.py
128 lines (101 loc) · 5.24 KB
/
colorizer_nongan.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
import os
import keras
from keras.layers import Conv2D ,UpSampling2D,Input,BatchNormalization, Activation, Dropout,Concatenate,MaxPooling2D
from keras.models import Model
from keras.optimizers import Adam
from keras.layers import LeakyReLU
import utils
import numpy as np
from keras.callbacks import ModelCheckpoint,CSVLogger,TensorBoard
from subpixel import SubpixelConv2D
import keras.backend as K
import argparse
print(K.image_dim_ordering())
## Модель глубокой нейронной сети на базе автоэнкодера.
## Попытка обучить нейронную сеть способную повышать пространственное разрешение
## спутниковых изображений.
def model_multi_input():
"""
Подготовка объекта класса Model, содержащего модель нейронной сети.
"""
inputs = Input(shape=(128,128,3))
conv1 = Conv2D(32, kernel_size=(3,3), padding='same')(inputs)
conv1 = BatchNormalization(momentum=0.8)(conv1)
conv1 = (LeakyReLU(0.2))(conv1)
conv1 = Conv2D(32, kernel_size=(3,3), padding='same')(conv1)
conv1 = BatchNormalization(momentum=0.8)(conv1)
conv1 = (LeakyReLU(0.2))(conv1)
conv2 = Conv2D(64, kernel_size=(3,3), padding='same')(conv1)
conv2 = BatchNormalization(momentum=0.8)(conv2)
conv2 = (LeakyReLU(0.2))(conv2)
conv2 = Conv2D(64, kernel_size=(3,3), strides=(2,2), padding='same')(conv2)
conv2 = BatchNormalization(momentum=0.8)(conv2)
conv2 = (LeakyReLU(0.2))(conv2)
conv3 = Conv2D(128, kernel_size=(3,3), padding='same')(conv2)
conv3 = BatchNormalization(momentum=0.8)(conv3)
conv3 = (LeakyReLU(0.2))(conv3)
conv3 = Conv2D(128, kernel_size=(3,3), strides=(2,2),padding='same')(conv3)
conv3 = BatchNormalization(momentum=0.8)(conv3)
conv3 = (LeakyReLU(0.2))(conv3)
conv4 = Conv2D(256, kernel_size=(3,3), padding='same')(conv3)
conv4 = BatchNormalization(momentum=0.8)(conv4)
conv4 = (LeakyReLU(0.2))(conv4)
conv4 = Conv2D(256, kernel_size=(3,3), strides=(2,2), padding='same')(conv4)
conv4 = BatchNormalization(momentum=0.8)(conv4)
conv4 = (LeakyReLU(0.2))(conv4)
up7 = Concatenate(axis=-1)([UpSampling2D(size=(2, 2))(conv4), conv3])
conv7 = Conv2D(128, kernel_size=(3,3), padding='same')(up7)
conv7 = BatchNormalization(momentum=0.8)(conv7)
conv7 = Activation('relu')(conv7)
conv7 = Conv2D(128, kernel_size=(3,3), padding='same')(conv7)
conv7 = BatchNormalization(momentum=0.8)(conv7)
conv7 = Activation('relu')(conv7)
up8 = Concatenate(axis=-1)([UpSampling2D(size=(2, 2))(conv7), conv2])
conv8 = Conv2D(64, kernel_size=(3,3), padding='same')(up8)
conv8 = BatchNormalization(momentum=0.8)(conv8)
conv8 = Activation('relu')(conv8)
conv8 = Conv2D(64, kernel_size=(3,3), padding='same')(conv8)
conv8 = BatchNormalization(momentum=0.8)(conv8)
conv8 = Activation('relu')(conv8)
up9 = Concatenate(axis=-1)([UpSampling2D(size=(2, 2))(conv8), conv1])
conv9 = Conv2D(32, kernel_size=(3,3), padding='same')(up9)
conv9 = BatchNormalization(momentum=0.8)(conv9)
conv9 = Activation('relu')(conv9)
conv9 = Conv2D(32, kernel_size=(3,3), padding='same')(conv9)
conv9 = BatchNormalization(momentum=0.8)(conv9)
conv9 = Activation('relu')(conv9)
conv10 = Conv2D(1,1,activation='sigmoid')(UpSampling2D(size=(2, 2))(conv9))
model = Model(input=inputs, output=conv10)
print('model compile')
return model
def main(args):
"""Основная функция, выполняющая обучение модели."""
# Количество эпох обучения.
num_steps = 1000
# Размер батча.
batch_size = 1
colorizer = model_multi_input()
if os.path.exists(args.model_path):
colorizer.load_weights(args.model_path)
print(colorizer.summary())
colorizer.compile(optimizer=Adam(lr=2e-4,beta_1=0.5), loss='binary_crossentropy', metrics=['mse','mae','accuracy'])
# Подготовка батчей.
igen = utils.batch_generator(path_x=args.data_path,batch_size=batch_size,max_steps=num_steps)
valx,valy = utils.load_single_pair(args.data_path,-200,100)
valx = np.array(valx)
valy = np.array(valy)
vali = (valx,valy)
print(np.histogram(valx))
print(np.histogram(valy))
# Настройка процесса обучения. Сохранение чекпоинтов, написание логов, "живая" визуалищация.
checkpoint = ModelCheckpoint(args.model_path, monitor='val_loss', verbose=1, save_best_only=True)
logger = CSVLogger(filename=args.model_path.replace('.mdl','.csv'), append=False)
tboard = TensorBoard(log_dir='./logs', histogram_freq=1, batch_size=1, write_graph=True, write_grads=True, write_images=True)
colorizer.fit_generator(igen,validation_data=vali, steps_per_epoch=250, epochs=10, verbose=1, callbacks=[checkpoint,logger,tboard],
max_queue_size=10,shuffle=True, initial_epoch=0,use_multiprocessing=False)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--data_path", required=True)
parser.add_argument("--model_path", required=True)
args = parser.parse_args()
main(args)